Merge pull request #74 from CJackHwang/codex/fix-toolcall-whitelist-issue

Recognize and emit executable tool_calls in mixed prose streams; normalize roles and loosen tool-name matching

API.md

@@ -284,6 +284,11 @@ data: [DONE]
 
 **流式**：命中高置信特征后立即输出 `delta.tool_calls`（不等待完整 JSON 闭合），并持续发送 arguments 增量；已确认的 toolcall 原始 JSON 不会回流到 `delta.content`。
 
+补充说明：
+
+- **非代码块上下文**下，工具 JSON 即使与普通文本混合，也会按特征识别并产出可执行 tool call（前后普通文本仍可透传）。
+- Markdown fenced code block（例如 ```json ... ```）中的 `tool_calls` 仅视为示例文本，不会被执行。
+
 ---
 
 ### `GET /v1/models/{id}`
@@ -301,7 +306,7 @@ OpenAI Responses 风格接口，兼容 `input` 或 `messages`。
 | `messages` | array | ❌ | 与 `input` 二选一 |
 | `instructions` | string | ❌ | 自动前置为 system 消息 |
 | `stream` | boolean | ❌ | 默认 `false` |
-| `tools` | array | ❌ | 与 chat 同样的工具识别与转译策略 |
+| `tools` | array | ❌ | 与 chat 同样的工具识别与转译策略（含代码块示例豁免） |
 | `tool_choice` | string/object | ❌ | 支持 `auto`/`none`/`required` 与强制函数（`{"type":"function","name":"..."}`） |
 
 **非流式响应**：返回标准 `response` 对象，`id` 形如 `resp_xxx`，并写入内存 TTL 存储。

README.en.md

@@ -351,6 +351,7 @@ Queue limit = DS2API_ACCOUNT_MAX_QUEUE (default = recommended concurrency)
 When `tools` is present in the request, DS2API performs anti-leak handling:
 
 1. Toolcall feature matching is enabled only in **non-code-block context** (fenced examples are ignored)
+   - In non-code-block context, tool JSON may still be recognized even when mixed with normal prose; surrounding prose can remain as text output.
 2. `responses` streaming strictly uses official item lifecycle events (`response.output_item.*`, `response.content_part.*`, `response.function_call_arguments.*`)
 3. Tool names not declared in the `tools` schema are strictly rejected and will not be emitted as valid tool calls
 4. `responses` supports and enforces `tool_choice` (`auto`/`none`/`required`/forced function); `required` violations return `422` for non-stream and `response.failed` for stream

TESTING.md

@@ -51,7 +51,7 @@ DS2API 提供两个层级的测试：
 1. **Preflight 检查**：
    - `go test ./... -count=1`（单元测试）
    - `./tests/scripts/check-node-split-syntax.sh`（Node 拆分模块语法门禁）
-   - `node --test api/helpers/stream-tool-sieve.test.js api/chat-stream.test.js api/compat/js_compat_test.js`（Node 流式拦截 + compat 单测）
+   - `node --test`（如仓库存在 Node 单测文件时执行；当前默认以 Go 测试 + Node 语法门禁为主）
    - `npm run build --prefix webui`（WebUI 构建检查）
 
 2. **隔离启动**：复制 `config.json` 到临时目录，启动独立服务进程

internal/adapter/gemini/handler_test.go

@@ -99,7 +99,7 @@ func TestGeminiRoutesRegistered(t *testing.T) {
 
 func TestGenerateContentReturnsFunctionCallParts(t *testing.T) {
 	upstream := makeGeminiUpstreamResponse(
-		`data: {"p":"response/content","v":"我来调用工具\n{\"tool_calls\":[{\"name\":\"eval_javascript\",\"input\":{\"code\":\"1+1\"}}]}"}`,
+		`data: {"p":"response/content","v":"{\"tool_calls\":[{\"name\":\"eval_javascript\",\"input\":{\"code\":\"1+1\"}}]}"}`,
 		`data: [DONE]`,
 	)
 	h := &Handler{
@@ -143,6 +143,42 @@ func TestGenerateContentReturnsFunctionCallParts(t *testing.T) {
 	}
 }
 
+func TestGenerateContentMixedToolSnippetAlsoTriggersFunctionCall(t *testing.T) {
+	upstream := makeGeminiUpstreamResponse(
+		`data: {"p":"response/content","v":"我来调用工具\n{\"tool_calls\":[{\"name\":\"eval_javascript\",\"input\":{\"code\":\"1+1\"}}]}"}`,
+		`data: [DONE]`,
+	)
+	h := &Handler{Store: testGeminiConfig{}, Auth: testGeminiAuth{}, DS: testGeminiDS{resp: upstream}}
+	r := chi.NewRouter()
+	RegisterRoutes(r, h)
+
+	body := `{
+		"contents":[{"role":"user","parts":[{"text":"call tool"}]}],
+		"tools":[{"functionDeclarations":[{"name":"eval_javascript","description":"eval","parameters":{"type":"object","properties":{"code":{"type":"string"}}}}]}]
+	}`
+	req := httptest.NewRequest(http.MethodPost, "/v1beta/models/gemini-2.5-pro:generateContent", strings.NewReader(body))
+	req.Header.Set("Authorization", "Bearer direct-token")
+	rec := httptest.NewRecorder()
+	r.ServeHTTP(rec, req)
+
+	if rec.Code != http.StatusOK {
+		t.Fatalf("expected 200, got %d body=%s", rec.Code, rec.Body.String())
+	}
+	var out map[string]any
+	if err := json.Unmarshal(rec.Body.Bytes(), &out); err != nil {
+		t.Fatalf("decode response failed: %v", err)
+	}
+	candidates, _ := out["candidates"].([]any)
+	c0, _ := candidates[0].(map[string]any)
+	content, _ := c0["content"].(map[string]any)
+	parts, _ := content["parts"].([]any)
+	part0, _ := parts[0].(map[string]any)
+	functionCall, _ := part0["functionCall"].(map[string]any)
+	if functionCall["name"] != "eval_javascript" {
+		t.Fatalf("expected functionCall name eval_javascript for mixed snippet, got %#v", functionCall)
+	}
+}
+
 func TestStreamGenerateContentEmitsSSE(t *testing.T) {
 	upstream := makeGeminiUpstreamResponse(
 		`data: {"p":"response/content","v":"hello "}`,

internal/adapter/openai/handler_toolcall_test.go

@@ -513,8 +513,8 @@ func TestHandleStreamToolCallMixedWithPlainTextSegments(t *testing.T) {
 	if !done {
 		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
 	}
-	if streamHasToolCallsDelta(frames) {
-		t.Fatalf("did not expect tool_calls delta in mixed prose stream, body=%s", rec.Body.String())
+	if !streamHasToolCallsDelta(frames) {
+		t.Fatalf("expected tool_calls delta in mixed prose stream, body=%s", rec.Body.String())
 	}
 	content := strings.Builder{}
 	for _, frame := range frames {
@@ -531,11 +531,8 @@ func TestHandleStreamToolCallMixedWithPlainTextSegments(t *testing.T) {
 	if !strings.Contains(got, "下面是示例：") || !strings.Contains(got, "请勿执行。") {
 		t.Fatalf("expected pre/post plain text to pass sieve, got=%q", got)
 	}
-	if !strings.Contains(strings.ToLower(got), `"tool_calls"`) {
-		t.Fatalf("expected embedded tool json to remain text in strict mode, got=%q", got)
-	}
-	if streamFinishReason(frames) != "stop" {
-		t.Fatalf("expected finish_reason=stop for mixed prose, body=%s", rec.Body.String())
+	if streamFinishReason(frames) != "tool_calls" {
+		t.Fatalf("expected finish_reason=tool_calls for mixed prose, body=%s", rec.Body.String())
 	}
 }
 
@@ -555,8 +552,8 @@ func TestHandleStreamToolCallAfterLeadingTextRemainsText(t *testing.T) {
 	if !done {
 		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
 	}
-	if streamHasToolCallsDelta(frames) {
-		t.Fatalf("did not expect tool_calls delta, body=%s", rec.Body.String())
+	if !streamHasToolCallsDelta(frames) {
+		t.Fatalf("expected tool_calls delta, body=%s", rec.Body.String())
 	}
 	content := strings.Builder{}
 	for _, frame := range frames {
@@ -573,11 +570,9 @@ func TestHandleStreamToolCallAfterLeadingTextRemainsText(t *testing.T) {
 	if !strings.Contains(got, "我将调用工具。") {
 		t.Fatalf("expected leading text to keep streaming, got=%q", got)
 	}
-	if !strings.Contains(strings.ToLower(got), "tool_calls") {
-		t.Fatalf("expected tool_calls example text preserved, got=%q", got)
-	}
-	if streamFinishReason(frames) != "stop" {
-		t.Fatalf("expected finish_reason=stop, body=%s", rec.Body.String())
+
+	if streamFinishReason(frames) != "tool_calls" {
+		t.Fatalf("expected finish_reason=tool_calls, body=%s", rec.Body.String())
 	}
 }
 
@@ -596,8 +591,8 @@ func TestHandleStreamToolCallWithSameChunkTrailingTextRemainsText(t *testing.T)
 	if !done {
 		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
 	}
-	if streamHasToolCallsDelta(frames) {
-		t.Fatalf("did not expect tool_calls delta, body=%s", rec.Body.String())
+	if !streamHasToolCallsDelta(frames) {
+		t.Fatalf("expected tool_calls delta, body=%s", rec.Body.String())
 	}
 	content := strings.Builder{}
 	for _, frame := range frames {
@@ -614,8 +609,45 @@ func TestHandleStreamToolCallWithSameChunkTrailingTextRemainsText(t *testing.T)
 	if !strings.Contains(got, "接下来我会继续说明。") {
 		t.Fatalf("expected trailing plain text to be preserved, got=%q", got)
 	}
-	if !strings.Contains(strings.ToLower(got), "tool_calls") {
-		t.Fatalf("expected tool_calls example text preserved, got=%q", got)
+
+	if streamFinishReason(frames) != "tool_calls" {
+		t.Fatalf("expected finish_reason=tool_calls, body=%s", rec.Body.String())
+	}
+}
+
+func TestHandleStreamFencedToolCallSnippetRemainsText(t *testing.T) {
+	h := &Handler{}
+	resp := makeSSEHTTPResponse(
+		fmt.Sprintf(`data: {"p":"response/content","v":%q}`, "下面是调用示例：\n```json\n"),
+		fmt.Sprintf(`data: {"p":"response/content","v":%q}`, "{\"tool_calls\":[{\"name\":\"search\",\"input\":{\"q\":\"go\"}}]}\n```\n仅示例，不要执行。"),
+		`data: [DONE]`,
+	)
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
+
+	h.handleStream(rec, req, resp, "cid7f", "deepseek-chat", "prompt", false, false, []string{"search"})
+
+	frames, done := parseSSEDataFrames(t, rec.Body.String())
+	if !done {
+		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
+	}
+	if streamHasToolCallsDelta(frames) {
+		t.Fatalf("did not expect tool_calls delta for fenced snippet, body=%s", rec.Body.String())
+	}
+	content := strings.Builder{}
+	for _, frame := range frames {
+		choices, _ := frame["choices"].([]any)
+		for _, item := range choices {
+			choice, _ := item.(map[string]any)
+			delta, _ := choice["delta"].(map[string]any)
+			if c, ok := delta["content"].(string); ok {
+				content.WriteString(c)
+			}
+		}
+	}
+	got := content.String()
+	if !strings.Contains(got, "```json") || !strings.Contains(strings.ToLower(got), "tool_calls") {
+		t.Fatalf("expected fenced tool snippet in content, got=%q", got)
 	}
 	if streamFinishReason(frames) != "stop" {
 		t.Fatalf("expected finish_reason=stop, body=%s", rec.Body.String())
@@ -640,8 +672,8 @@ func TestHandleStreamToolCallKeyAppearsLateRemainsText(t *testing.T) {
 	if !done {
 		t.Fatalf("expected [DONE], body=%s", rec.Body.String())
 	}
-	if streamHasToolCallsDelta(frames) {
-		t.Fatalf("did not expect tool_calls delta, body=%s", rec.Body.String())
+	if !streamHasToolCallsDelta(frames) {
+		t.Fatalf("expected tool_calls delta, body=%s", rec.Body.String())
 	}
 	content := strings.Builder{}
 	for _, frame := range frames {
@@ -655,14 +687,11 @@ func TestHandleStreamToolCallKeyAppearsLateRemainsText(t *testing.T) {
 		}
 	}
 	got := content.String()
-	if !strings.Contains(strings.ToLower(got), "tool_calls") || !strings.Contains(got, "{") {
-		t.Fatalf("expected embedded tool json to remain in text, got=%q", got)
-	}
 	if !strings.Contains(got, "后置正文C。") {
 		t.Fatalf("expected stream to continue after tool json convergence, got=%q", got)
 	}
-	if streamFinishReason(frames) != "stop" {
-		t.Fatalf("expected finish_reason=stop, body=%s", rec.Body.String())
+	if streamFinishReason(frames) != "tool_calls" {
+		t.Fatalf("expected finish_reason=tool_calls, body=%s", rec.Body.String())
 	}
 }
 

internal/adapter/openai/message_normalize.go

@@ -33,9 +33,9 @@ func normalizeOpenAIMessagesForPrompt(raw []any, traceID string) []map[string]an
 				"role":    "user",
 				"content": formatToolResultForPrompt(msg),
 			})
-		case "user", "system":
+		case "user", "system", "developer":
 			out = append(out, map[string]any{
-				"role":    role,
+				"role":    normalizeOpenAIRoleForPrompt(role),
 				"content": normalizeOpenAIContentForPrompt(msg["content"]),
 			})
 		default:
@@ -47,7 +47,7 @@ func normalizeOpenAIMessagesForPrompt(raw []any, traceID string) []map[string]an
 				role = "user"
 			}
 			out = append(out, map[string]any{
-				"role":    role,
+				"role":    normalizeOpenAIRoleForPrompt(role),
 				"content": content,
 			})
 		}
@@ -189,6 +189,14 @@ func marshalToPromptString(v any) string {
 	return string(b)
 }
 
+func normalizeOpenAIRoleForPrompt(role string) string {
+	role = strings.ToLower(strings.TrimSpace(role))
+	if role == "developer" {
+		return "system"
+	}
+	return role
+}
+
 func asString(v any) string {
 	if s, ok := v.(string); ok {
 		return s

internal/adapter/openai/message_normalize_test.go

@@ -193,3 +193,17 @@ func TestNormalizeOpenAIMessagesForPrompt_PreservesConcatenatedToolArguments(t *
 		t.Fatalf("expected original concatenated arguments in tool history, got %q", content)
 	}
 }
+
+func TestNormalizeOpenAIMessagesForPrompt_DeveloperRoleMapsToSystem(t *testing.T) {
+	raw := []any{
+		map[string]any{"role": "developer", "content": "必须先走工具调用"},
+		map[string]any{"role": "user", "content": "你好"},
+	}
+	normalized := normalizeOpenAIMessagesForPrompt(raw, "")
+	if len(normalized) != 2 {
+		t.Fatalf("expected 2 normalized messages, got %d", len(normalized))
+	}
+	if normalized[0]["role"] != "system" {
+		t.Fatalf("expected developer role converted to system, got %#v", normalized[0]["role"])
+	}
+}

internal/adapter/openai/responses_input_items.go

@@ -29,7 +29,7 @@ func normalizeResponsesInputItemWithState(m map[string]any, callNameByID map[str
 			return nil
 		}
 		return map[string]any{
-			"role":    role,
+			"role":    normalizeOpenAIRoleForPrompt(role),
 			"content": content,
 		}
 	}
@@ -51,7 +51,7 @@ func normalizeResponsesInputItemWithState(m map[string]any, callNameByID map[str
 			role = "user"
 		}
 		return map[string]any{
-			"role":    role,
+			"role":    normalizeOpenAIRoleForPrompt(role),
 			"content": content,
 		}
 	case "function_call_output", "tool_result":

internal/adapter/openai/responses_stream_test.go

@@ -288,12 +288,8 @@ func TestHandleResponsesStreamThinkingAndMixedToolExampleRemainMessageOnly(t *te
 	h.handleResponsesStream(rec, req, resp, "owner-a", "resp_test", "deepseek-reasoner", "prompt", true, false, []string{"read_file"}, util.DefaultToolChoicePolicy(), "")
 
 	addedPayloads := extractAllSSEEventPayloads(rec.Body.String(), "response.output_item.added")
-	if len(addedPayloads) != 1 {
-		t.Fatalf("expected only one message output_item.added event, got %d body=%s", len(addedPayloads), rec.Body.String())
-	}
-	item, _ := addedPayloads[0]["item"].(map[string]any)
-	if asString(item["type"]) != "message" {
-		t.Fatalf("expected only message output item in strict mode, got %#v", item)
+	if len(addedPayloads) < 1 {
+		t.Fatalf("expected at least one output_item.added event, got %d body=%s", len(addedPayloads), rec.Body.String())
 	}
 
 	completedPayload, ok := extractSSEEventPayload(rec.Body.String(), "response.completed")
@@ -302,15 +298,22 @@ func TestHandleResponsesStreamThinkingAndMixedToolExampleRemainMessageOnly(t *te
 	}
 	responseObj, _ := completedPayload["response"].(map[string]any)
 	output, _ := responseObj["output"].([]any)
+	hasMessage := false
 	for _, item := range output {
 		m, _ := item.(map[string]any)
 		if m == nil {
 			continue
 		}
+		if asString(m["type"]) == "message" {
+			hasMessage = true
+		}
 		if asString(m["type"]) == "function_call" {
 			t.Fatalf("did not expect function_call output for mixed prose tool example, output=%#v", output)
 		}
 	}
+	if !hasMessage {
+		t.Fatalf("expected message output for mixed prose tool example, output=%#v", output)
+	}
 }
 
 func TestHandleResponsesStreamToolChoiceNoneRejectsFunctionCall(t *testing.T) {

internal/adapter/openai/tool_sieve_core.go

@@ -15,19 +15,9 @@ func processToolSieveChunk(state *toolStreamSieveState, chunk string, toolNames
 	}
 	events := make([]toolStreamEvent, 0, 2)
 	if len(state.pendingToolCalls) > 0 {
-		pending := state.pending.String()
-		if strings.TrimSpace(pending) != "" {
-			content := state.pendingToolRaw + pending
-			state.pending.Reset()
-			state.pendingToolRaw = ""
-			state.pendingToolCalls = nil
-			state.noteText(content)
-			events = append(events, toolStreamEvent{Content: content})
-		} else {
-			// Wait for either more non-whitespace content (demote to plain text)
-			// or stream flush (promote to executable tool calls).
-			return events
-		}
+		events = append(events, toolStreamEvent{ToolCalls: state.pendingToolCalls})
+		state.pendingToolRaw = ""
+		state.pendingToolCalls = nil
 	}
 
 	for {
@@ -45,7 +35,14 @@ func processToolSieveChunk(state *toolStreamSieveState, chunk string, toolNames
 			state.capturing = false
 			state.resetIncrementalToolState()
 			if len(calls) > 0 {
-				state.pendingToolRaw = captured
+				if prefix != "" {
+					state.noteText(prefix)
+					events = append(events, toolStreamEvent{Content: prefix})
+				}
+				if suffix != "" {
+					state.pending.WriteString(suffix)
+				}
+				_ = captured
 				state.pendingToolCalls = calls
 				continue
 			}
@@ -211,11 +208,6 @@ func consumeToolCapture(state *toolStreamSieveState, toolNames []string) (prefix
 	if insideCodeFence(state.recentTextTail + prefixPart) {
 		return captured, nil, "", true
 	}
-	// Strict mode: only standalone tool payloads are executable. If the
-	// payload is wrapped by non-whitespace prose, keep it as plain text.
-	if strings.TrimSpace(state.recentTextTail) != "" || strings.TrimSpace(prefixPart) != "" || strings.TrimSpace(suffixPart) != "" {
-		return captured, nil, "", true
-	}
 	parsed := util.ParseStandaloneToolCallsDetailed(obj, toolNames)
 	if len(parsed.Calls) == 0 {
 		if parsed.SawToolCallSyntax && parsed.RejectedByPolicy {

internal/admin/handler_accounts_testing.go

@@ -1,209 +1,212 @@
-package admin
-
-import (
-	"bytes"
-	"context"
-	"encoding/json"
-	"fmt"
-	"io"
-	"net/http"
-	"strings"
-	"sync"
-	"time"
-
-	authn "ds2api/internal/auth"
-	"ds2api/internal/config"
-	"ds2api/internal/sse"
-)
-
-func (h *Handler) testSingleAccount(w http.ResponseWriter, r *http.Request) {
-	var req map[string]any
-	_ = json.NewDecoder(r.Body).Decode(&req)
-	identifier, _ := req["identifier"].(string)
-	if strings.TrimSpace(identifier) == "" {
-		writeJSON(w, http.StatusBadRequest, map[string]any{"detail": "需要账号标识（identifier / email / mobile）"})
-		return
-	}
-	acc, ok := findAccountByIdentifier(h.Store, identifier)
-	if !ok {
-		writeJSON(w, http.StatusNotFound, map[string]any{"detail": "账号不存在"})
-		return
-	}
-	model, _ := req["model"].(string)
-	if model == "" {
-		model = "deepseek-chat"
-	}
-	message, _ := req["message"].(string)
-	result := h.testAccount(r.Context(), acc, model, message)
-	writeJSON(w, http.StatusOK, result)
-}
-
-func (h *Handler) testAllAccounts(w http.ResponseWriter, r *http.Request) {
-	var req map[string]any
-	_ = json.NewDecoder(r.Body).Decode(&req)
-	model, _ := req["model"].(string)
-	if model == "" {
-		model = "deepseek-chat"
-	}
-	accounts := h.Store.Snapshot().Accounts
-	if len(accounts) == 0 {
-		writeJSON(w, http.StatusOK, map[string]any{"total": 0, "success": 0, "failed": 0, "results": []any{}})
-		return
-	}
-
-	// Concurrent testing with a semaphore to limit parallelism.
-	const maxConcurrency = 5
-	results := runAccountTestsConcurrently(accounts, maxConcurrency, func(_ int, account config.Account) map[string]any {
-		return h.testAccount(r.Context(), account, model, "")
-	})
-
-	success := 0
-	for _, res := range results {
-		if ok, _ := res["success"].(bool); ok {
-			success++
-		}
-	}
-	writeJSON(w, http.StatusOK, map[string]any{"total": len(accounts), "success": success, "failed": len(accounts) - success, "results": results})
-}
-
-func runAccountTestsConcurrently(accounts []config.Account, maxConcurrency int, testFn func(int, config.Account) map[string]any) []map[string]any {
-	if maxConcurrency <= 0 {
-		maxConcurrency = 1
-	}
-	sem := make(chan struct{}, maxConcurrency)
-	results := make([]map[string]any, len(accounts))
-	var wg sync.WaitGroup
-	for i, acc := range accounts {
-		wg.Add(1)
-		go func(idx int, account config.Account) {
-			defer wg.Done()
-			sem <- struct{}{}        // acquire
-			defer func() { <-sem }() // release
-			results[idx] = testFn(idx, account)
-		}(i, acc)
-	}
-	wg.Wait()
-	return results
-}
-
-func (h *Handler) testAccount(ctx context.Context, acc config.Account, model, message string) map[string]any {
-	start := time.Now()
-	identifier := acc.Identifier()
-	result := map[string]any{"account": identifier, "success": false, "response_time": 0, "message": "", "model": model}
-	defer func() {
-		status := "failed"
-		if ok, _ := result["success"].(bool); ok {
-			status = "ok"
-		}
-		_ = h.Store.UpdateAccountTestStatus(identifier, status)
-	}()
-	token := strings.TrimSpace(acc.Token)
-	if token == "" {
-		newToken, err := h.DS.Login(ctx, acc)
-		if err != nil {
-			result["message"] = "登录失败: " + err.Error()
-			return result
-		}
-		token = newToken
-		_ = h.Store.UpdateAccountToken(acc.Identifier(), token)
-	}
-	authCtx := &authn.RequestAuth{UseConfigToken: false, DeepSeekToken: token}
-	sessionID, err := h.DS.CreateSession(ctx, authCtx, 1)
-	if err != nil {
-		newToken, loginErr := h.DS.Login(ctx, acc)
-		if loginErr != nil {
-			result["message"] = "创建会话失败: " + err.Error()
-			return result
-		}
-		token = newToken
-		authCtx.DeepSeekToken = token
-		_ = h.Store.UpdateAccountToken(acc.Identifier(), token)
-		sessionID, err = h.DS.CreateSession(ctx, authCtx, 1)
-		if err != nil {
-			result["message"] = "创建会话失败: " + err.Error()
-			return result
-		}
-	}
-	if strings.TrimSpace(message) == "" {
-		message = "你是谁？"
-	}
-	thinking, search, ok := config.GetModelConfig(model)
-	if !ok {
-		thinking, search = false, false
-	}
-	_ = search
-	pow, err := h.DS.GetPow(ctx, authCtx, 1)
-	if err != nil {
-		result["message"] = "获取 PoW 失败: " + err.Error()
-		return result
-	}
-	payload := map[string]any{"chat_session_id": sessionID, "prompt": "<｜User｜>" + message, "ref_file_ids": []any{}, "thinking_enabled": thinking, "search_enabled": search}
-	resp, err := h.DS.CallCompletion(ctx, authCtx, payload, pow, 1)
-	if err != nil {
-		result["message"] = "请求失败: " + err.Error()
-		return result
-	}
-	if resp.StatusCode != http.StatusOK {
-		defer resp.Body.Close()
-		result["message"] = fmt.Sprintf("请求失败: HTTP %d", resp.StatusCode)
-		return result
-	}
-	collected := sse.CollectStream(resp, thinking, true)
-	result["success"] = true
-	result["response_time"] = int(time.Since(start).Milliseconds())
-	if collected.Text != "" {
-		result["message"] = collected.Text
-	} else {
-		result["message"] = "（无回复内容）"
-	}
-	if collected.Thinking != "" {
-		result["thinking"] = collected.Thinking
-	}
-	return result
-}
-
-func (h *Handler) testAPI(w http.ResponseWriter, r *http.Request) {
-	var req map[string]any
-	_ = json.NewDecoder(r.Body).Decode(&req)
-	model, _ := req["model"].(string)
-	message, _ := req["message"].(string)
-	apiKey, _ := req["api_key"].(string)
-	if model == "" {
-		model = "deepseek-chat"
-	}
-	if message == "" {
-		message = "你好"
-	}
-	if apiKey == "" {
-		keys := h.Store.Snapshot().Keys
-		if len(keys) == 0 {
-			writeJSON(w, http.StatusBadRequest, map[string]any{"detail": "没有可用的 API Key"})
-			return
-		}
-		apiKey = keys[0]
-	}
-	host := r.Host
-	scheme := "http"
-	if strings.Contains(strings.ToLower(host), "vercel") || strings.Contains(strings.ToLower(r.Header.Get("X-Forwarded-Proto")), "https") {
-		scheme = "https"
-	}
-	payload := map[string]any{"model": model, "messages": []map[string]any{{"role": "user", "content": message}}, "stream": false}
-	b, _ := json.Marshal(payload)
-	request, _ := http.NewRequestWithContext(r.Context(), http.MethodPost, fmt.Sprintf("%s://%s/v1/chat/completions", scheme, host), bytes.NewReader(b))
-	request.Header.Set("Authorization", "Bearer "+apiKey)
-	request.Header.Set("Content-Type", "application/json")
-	resp, err := (&http.Client{Timeout: 60 * time.Second}).Do(request)
-	if err != nil {
-		writeJSON(w, http.StatusOK, map[string]any{"success": false, "error": err.Error()})
-		return
-	}
-	defer resp.Body.Close()
-	body, _ := io.ReadAll(resp.Body)
-	if resp.StatusCode == http.StatusOK {
-		var parsed any
-		_ = json.Unmarshal(body, &parsed)
-		writeJSON(w, http.StatusOK, map[string]any{"success": true, "status_code": resp.StatusCode, "response": parsed})
-		return
-	}
-	writeJSON(w, http.StatusOK, map[string]any{"success": false, "status_code": resp.StatusCode, "response": string(body)})
-}
+package admin
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"sync"
+	"time"
+
+	authn "ds2api/internal/auth"
+	"ds2api/internal/config"
+	"ds2api/internal/sse"
+)
+
+func (h *Handler) testSingleAccount(w http.ResponseWriter, r *http.Request) {
+	var req map[string]any
+	_ = json.NewDecoder(r.Body).Decode(&req)
+	identifier, _ := req["identifier"].(string)
+	if strings.TrimSpace(identifier) == "" {
+		writeJSON(w, http.StatusBadRequest, map[string]any{"detail": "需要账号标识（identifier / email / mobile）"})
+		return
+	}
+	acc, ok := findAccountByIdentifier(h.Store, identifier)
+	if !ok {
+		writeJSON(w, http.StatusNotFound, map[string]any{"detail": "账号不存在"})
+		return
+	}
+	model, _ := req["model"].(string)
+	if model == "" {
+		model = "deepseek-chat"
+	}
+	message, _ := req["message"].(string)
+	result := h.testAccount(r.Context(), acc, model, message)
+	writeJSON(w, http.StatusOK, result)
+}
+
+func (h *Handler) testAllAccounts(w http.ResponseWriter, r *http.Request) {
+	var req map[string]any
+	_ = json.NewDecoder(r.Body).Decode(&req)
+	model, _ := req["model"].(string)
+	if model == "" {
+		model = "deepseek-chat"
+	}
+	accounts := h.Store.Snapshot().Accounts
+	if len(accounts) == 0 {
+		writeJSON(w, http.StatusOK, map[string]any{"total": 0, "success": 0, "failed": 0, "results": []any{}})
+		return
+	}
+
+	// Concurrent testing with a semaphore to limit parallelism.
+	const maxConcurrency = 5
+	results := runAccountTestsConcurrently(accounts, maxConcurrency, func(_ int, account config.Account) map[string]any {
+		return h.testAccount(r.Context(), account, model, "")
+	})
+
+	success := 0
+	for _, res := range results {
+		if ok, _ := res["success"].(bool); ok {
+			success++
+		}
+	}
+	writeJSON(w, http.StatusOK, map[string]any{"total": len(accounts), "success": success, "failed": len(accounts) - success, "results": results})
+}
+
+func runAccountTestsConcurrently(accounts []config.Account, maxConcurrency int, testFn func(int, config.Account) map[string]any) []map[string]any {
+	if maxConcurrency <= 0 {
+		maxConcurrency = 1
+	}
+	sem := make(chan struct{}, maxConcurrency)
+	results := make([]map[string]any, len(accounts))
+	var wg sync.WaitGroup
+	for i, acc := range accounts {
+		wg.Add(1)
+		go func(idx int, account config.Account) {
+			defer wg.Done()
+			sem <- struct{}{}        // acquire
+			defer func() { <-sem }() // release
+			results[idx] = testFn(idx, account)
+		}(i, acc)
+	}
+	wg.Wait()
+	return results
+}
+
+func (h *Handler) testAccount(ctx context.Context, acc config.Account, model, message string) map[string]any {
+	start := time.Now()
+	identifier := acc.Identifier()
+	result := map[string]any{"account": identifier, "success": false, "response_time": 0, "message": "", "model": model}
+	defer func() {
+		status := "failed"
+		if ok, _ := result["success"].(bool); ok {
+			status = "ok"
+		}
+		_ = h.Store.UpdateAccountTestStatus(identifier, status)
+	}()
+	token := strings.TrimSpace(acc.Token)
+	if token == "" {
+		newToken, err := h.DS.Login(ctx, acc)
+		if err != nil {
+			result["message"] = "登录失败: " + err.Error()
+			return result
+		}
+		token = newToken
+		_ = h.Store.UpdateAccountToken(acc.Identifier(), token)
+	}
+	authCtx := &authn.RequestAuth{UseConfigToken: false, DeepSeekToken: token}
+	sessionID, err := h.DS.CreateSession(ctx, authCtx, 1)
+	if err != nil {
+		newToken, loginErr := h.DS.Login(ctx, acc)
+		if loginErr != nil {
+			result["message"] = "创建会话失败: " + err.Error()
+			return result
+		}
+		token = newToken
+		authCtx.DeepSeekToken = token
+		_ = h.Store.UpdateAccountToken(acc.Identifier(), token)
+		sessionID, err = h.DS.CreateSession(ctx, authCtx, 1)
+		if err != nil {
+			result["message"] = "创建会话失败: " + err.Error()
+			return result
+		}
+	}
+	if strings.TrimSpace(message) == "" {
+		result["success"] = true
+		result["message"] = "API 测试成功（仅会话创建）"
+		result["response_time"] = int(time.Since(start).Milliseconds())
+		return result
+	}
+	thinking, search, ok := config.GetModelConfig(model)
+	if !ok {
+		thinking, search = false, false
+	}
+	_ = search
+	pow, err := h.DS.GetPow(ctx, authCtx, 1)
+	if err != nil {
+		result["message"] = "获取 PoW 失败: " + err.Error()
+		return result
+	}
+	payload := map[string]any{"chat_session_id": sessionID, "prompt": "<｜User｜>" + message, "ref_file_ids": []any{}, "thinking_enabled": thinking, "search_enabled": search}
+	resp, err := h.DS.CallCompletion(ctx, authCtx, payload, pow, 1)
+	if err != nil {
+		result["message"] = "请求失败: " + err.Error()
+		return result
+	}
+	if resp.StatusCode != http.StatusOK {
+		defer resp.Body.Close()
+		result["message"] = fmt.Sprintf("请求失败: HTTP %d", resp.StatusCode)
+		return result
+	}
+	collected := sse.CollectStream(resp, thinking, true)
+	result["success"] = true
+	result["response_time"] = int(time.Since(start).Milliseconds())
+	if collected.Text != "" {
+		result["message"] = collected.Text
+	} else {
+		result["message"] = "（无回复内容）"
+	}
+	if collected.Thinking != "" {
+		result["thinking"] = collected.Thinking
+	}
+	return result
+}
+
+func (h *Handler) testAPI(w http.ResponseWriter, r *http.Request) {
+	var req map[string]any
+	_ = json.NewDecoder(r.Body).Decode(&req)
+	model, _ := req["model"].(string)
+	message, _ := req["message"].(string)
+	apiKey, _ := req["api_key"].(string)
+	if model == "" {
+		model = "deepseek-chat"
+	}
+	if message == "" {
+		message = "你好"
+	}
+	if apiKey == "" {
+		keys := h.Store.Snapshot().Keys
+		if len(keys) == 0 {
+			writeJSON(w, http.StatusBadRequest, map[string]any{"detail": "没有可用的 API Key"})
+			return
+		}
+		apiKey = keys[0]
+	}
+	host := r.Host
+	scheme := "http"
+	if strings.Contains(strings.ToLower(host), "vercel") || strings.Contains(strings.ToLower(r.Header.Get("X-Forwarded-Proto")), "https") {
+		scheme = "https"
+	}
+	payload := map[string]any{"model": model, "messages": []map[string]any{{"role": "user", "content": message}}, "stream": false}
+	b, _ := json.Marshal(payload)
+	request, _ := http.NewRequestWithContext(r.Context(), http.MethodPost, fmt.Sprintf("%s://%s/v1/chat/completions", scheme, host), bytes.NewReader(b))
+	request.Header.Set("Authorization", "Bearer "+apiKey)
+	request.Header.Set("Content-Type", "application/json")
+	resp, err := (&http.Client{Timeout: 60 * time.Second}).Do(request)
+	if err != nil {
+		writeJSON(w, http.StatusOK, map[string]any{"success": false, "error": err.Error()})
+		return
+	}
+	defer resp.Body.Close()
+	body, _ := io.ReadAll(resp.Body)
+	if resp.StatusCode == http.StatusOK {
+		var parsed any
+		_ = json.Unmarshal(body, &parsed)
+		writeJSON(w, http.StatusOK, map[string]any{"success": true, "status_code": resp.StatusCode, "response": parsed})
+		return
+	}
+	writeJSON(w, http.StatusOK, map[string]any{"success": false, "status_code": resp.StatusCode, "response": string(body)})
+}

internal/admin/handler_accounts_testing_test.go

@@ -0,0 +1,76 @@
+package admin
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"strings"
+	"testing"
+
+	"ds2api/internal/auth"
+	"ds2api/internal/config"
+)
+
+type testingDSMock struct {
+	loginCalls          int
+	createSessionCalls  int
+	getPowCalls         int
+	callCompletionCalls int
+}
+
+func (m *testingDSMock) Login(_ context.Context, _ config.Account) (string, error) {
+	m.loginCalls++
+	return "new-token", nil
+}
+
+func (m *testingDSMock) CreateSession(_ context.Context, _ *auth.RequestAuth, _ int) (string, error) {
+	m.createSessionCalls++
+	return "session-id", nil
+}
+
+func (m *testingDSMock) GetPow(_ context.Context, _ *auth.RequestAuth, _ int) (string, error) {
+	m.getPowCalls++
+	return "", errors.New("should not call GetPow in this test")
+}
+
+func (m *testingDSMock) CallCompletion(_ context.Context, _ *auth.RequestAuth, _ map[string]any, _ string, _ int) (*http.Response, error) {
+	m.callCompletionCalls++
+	return nil, errors.New("should not call CallCompletion in this test")
+}
+
+func TestTestAccount_BatchModeOnlyCreatesSession(t *testing.T) {
+	t.Setenv("DS2API_CONFIG_JSON", `{"accounts":[{"email":"batch@example.com","password":"pwd","token":""}]}`)
+	store := config.LoadStore()
+	ds := &testingDSMock{}
+	h := &Handler{Store: store, DS: ds}
+	acc, ok := store.FindAccount("batch@example.com")
+	if !ok {
+		t.Fatal("expected test account")
+	}
+
+	result := h.testAccount(context.Background(), acc, "deepseek-chat", "")
+
+	if ok, _ := result["success"].(bool); !ok {
+		t.Fatalf("expected success=true, got %#v", result)
+	}
+	msg, _ := result["message"].(string)
+	if !strings.Contains(msg, "仅会话创建") {
+		t.Fatalf("expected session-only success message, got %q", msg)
+	}
+	if ds.loginCalls != 1 || ds.createSessionCalls != 1 {
+		t.Fatalf("unexpected Login/CreateSession calls: login=%d createSession=%d", ds.loginCalls, ds.createSessionCalls)
+	}
+	if ds.getPowCalls != 0 || ds.callCompletionCalls != 0 {
+		t.Fatalf("expected no completion flow calls, got getPow=%d callCompletion=%d", ds.getPowCalls, ds.callCompletionCalls)
+	}
+	updated, ok := store.FindAccount("batch@example.com")
+	if !ok {
+		t.Fatal("expected updated account")
+	}
+	if updated.Token != "new-token" {
+		t.Fatalf("expected refreshed token to be persisted, got %q", updated.Token)
+	}
+	if updated.TestStatus != "ok" {
+		t.Fatalf("expected test status ok, got %q", updated.TestStatus)
+	}
+}

internal/deepseek/client_auth.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"net/http"
 	"strings"
+	"unicode"
 
 	"ds2api/internal/auth"
 	"ds2api/internal/config"
@@ -20,8 +21,9 @@ func (c *Client) Login(ctx context.Context, acc config.Account) (string, error)
 	if email := strings.TrimSpace(acc.Email); email != "" {
 		payload["email"] = email
 	} else if mobile := strings.TrimSpace(acc.Mobile); mobile != "" {
-		payload["mobile"] = mobile
-		payload["area_code"] = nil
+		loginMobile, areaCode := normalizeMobileForLogin(mobile)
+		payload["mobile"] = loginMobile
+		payload["area_code"] = areaCode
 	} else {
 		return "", errors.New("missing email/mobile")
 	}
@@ -151,3 +153,26 @@ func isTokenInvalid(status int, code int, msg string) bool {
 	}
 	return strings.Contains(msg, "token") || strings.Contains(msg, "unauthorized")
 }
+
+func normalizeMobileForLogin(raw string) (mobile string, areaCode any) {
+	s := strings.TrimSpace(raw)
+	if s == "" {
+		return "", nil
+	}
+	hasPlus := strings.HasPrefix(s, "+")
+	var b strings.Builder
+	b.Grow(len(s))
+	for _, r := range s {
+		if unicode.IsDigit(r) {
+			b.WriteRune(r)
+		}
+	}
+	digits := b.String()
+	if digits == "" {
+		return "", nil
+	}
+	if (hasPlus || strings.HasPrefix(digits, "86")) && strings.HasPrefix(digits, "86") && len(digits) == 13 {
+		return digits[2:], nil
+	}
+	return digits, nil
+}

internal/deepseek/client_auth_mobile_test.go

@@ -0,0 +1,33 @@
+package deepseek
+
+import "testing"
+
+func TestNormalizeMobileForLogin_ChinaWithPlus86(t *testing.T) {
+	mobile, areaCode := normalizeMobileForLogin("+8613800138000")
+	if mobile != "13800138000" {
+		t.Fatalf("unexpected mobile: %q", mobile)
+	}
+	if areaCode != nil {
+		t.Fatalf("expected nil areaCode, got %#v", areaCode)
+	}
+}
+
+func TestNormalizeMobileForLogin_ChinaWith86Prefix(t *testing.T) {
+	mobile, areaCode := normalizeMobileForLogin("8613800138000")
+	if mobile != "13800138000" {
+		t.Fatalf("unexpected mobile: %q", mobile)
+	}
+	if areaCode != nil {
+		t.Fatalf("expected nil areaCode, got %#v", areaCode)
+	}
+}
+
+func TestNormalizeMobileForLogin_KeepPlainDigits(t *testing.T) {
+	mobile, areaCode := normalizeMobileForLogin("13800138000")
+	if mobile != "13800138000" {
+		t.Fatalf("unexpected mobile: %q", mobile)
+	}
+	if areaCode != nil {
+		t.Fatalf("expected nil areaCode, got %#v", areaCode)
+	}
+}

internal/util/toolcalls_parse.go

@@ -2,9 +2,12 @@ package util
 
 import (
 	"encoding/json"
+	"regexp"
 	"strings"
 )
 
+var toolNameLoosePattern = regexp.MustCompile(`[^a-z0-9]+`)
+
 type ParsedToolCall struct {
 	Name  string         `json:"name"`
 	Input map[string]any `json:"input"`
@@ -121,12 +124,7 @@ func filterToolCallsDetailed(parsed []ParsedToolCall, availableToolNames []strin
 		if tc.Name == "" {
 			continue
 		}
-		matchedName := ""
-		if _, ok := allowed[tc.Name]; ok {
-			matchedName = tc.Name
-		} else if canonical, ok := allowedCanonical[strings.ToLower(tc.Name)]; ok {
-			matchedName = canonical
-		}
+		matchedName := resolveAllowedToolName(tc.Name, allowed, allowedCanonical)
 		if matchedName == "" {
 			rejectedSet[tc.Name] = struct{}{}
 			continue
@@ -144,6 +142,31 @@ func filterToolCallsDetailed(parsed []ParsedToolCall, availableToolNames []strin
 	return out, rejected
 }
 
+func resolveAllowedToolName(name string, allowed map[string]struct{}, allowedCanonical map[string]string) string {
+	if _, ok := allowed[name]; ok {
+		return name
+	}
+	lower := strings.ToLower(strings.TrimSpace(name))
+	if canonical, ok := allowedCanonical[lower]; ok {
+		return canonical
+	}
+	if idx := strings.LastIndex(lower, "."); idx >= 0 && idx < len(lower)-1 {
+		if canonical, ok := allowedCanonical[lower[idx+1:]]; ok {
+			return canonical
+		}
+	}
+	loose := toolNameLoosePattern.ReplaceAllString(lower, "")
+	if loose == "" {
+		return ""
+	}
+	for candidateLower, canonical := range allowedCanonical {
+		if toolNameLoosePattern.ReplaceAllString(candidateLower, "") == loose {
+			return canonical
+		}
+	}
+	return ""
+}
+
 func parseToolCallsPayload(payload string) []ParsedToolCall {
 	var decoded any
 	if err := json.Unmarshal([]byte(payload), &decoded); err != nil {

internal/util/toolcalls_test.go

@@ -115,3 +115,25 @@ func TestParseStandaloneToolCallsIgnoresFencedCodeBlock(t *testing.T) {
 		t.Fatalf("expected fenced tool_call example to be ignored, got %#v", calls)
 	}
 }
+
+func TestParseToolCallsAllowsQualifiedToolName(t *testing.T) {
+	text := `{"tool_calls":[{"name":"mcp.search_web","input":{"q":"golang"}}]}`
+	calls := ParseToolCalls(text, []string{"search_web"})
+	if len(calls) != 1 {
+		t.Fatalf("expected 1 call, got %#v", calls)
+	}
+	if calls[0].Name != "search_web" {
+		t.Fatalf("expected canonical tool name search_web, got %q", calls[0].Name)
+	}
+}
+
+func TestParseToolCallsAllowsPunctuationVariantToolName(t *testing.T) {
+	text := `{"tool_calls":[{"name":"read-file","input":{"path":"README.md"}}]}`
+	calls := ParseToolCalls(text, []string{"read_file"})
+	if len(calls) != 1 {
+		t.Fatalf("expected 1 call, got %#v", calls)
+	}
+	if calls[0].Name != "read_file" {
+		t.Fatalf("expected canonical tool name read_file, got %q", calls[0].Name)
+	}
+}